diff --git a/optimum/intel/openvino/modeling_visual_language.py b/optimum/intel/openvino/modeling_visual_language.py
index 141abeb87..86463b160 100644
--- a/optimum/intel/openvino/modeling_visual_language.py
+++ b/optimum/intel/openvino/modeling_visual_language.py
@@ -54,14 +54,25 @@ def __init__(
 
     def compile(self):
         if self.request is None:
-            logger.info(f"Compiling the Language model to {self._device} ...")
-            self.request = core.compile_model(self.model, self._device, self.ov_config).create_infer_request()
+            if self._compile_only:
+                self.request = self.model.create_infer_request()
+            else:
+                logger.info(f"Compiling the Language model to {self._device} ...")
+                self.request = self._compile_model(
+                    self.model, self._device, self.ov_config, self.model_save_dir
+                ).create_infer_request()
         self._compile_text_emb()
 
     def _compile_text_emb(self):
         if self.text_emb_request is None:
             logger.info(f"Compiling the Text embeddings model to {self._device} ...")
-            self.text_emb_request = core.compile_model(self.text_emb_model, self._device, self.ov_config)
+            if self._compile_only:
+                self.text_emb_request = self.text_emb_model
+            else:
+                logger.info(f"Compiling the Text embeddings model to {self._device} ...")
+                self.text_emb_request = self._compile_model(
+                    self.text_emb_model, self._device, self.ov_config, self.model_save_dir
+                )
 
     def clear_requests(self):
         if self._compile_only:
@@ -122,8 +133,8 @@ def prepare_inputs(
             else:
                 position_ids = np.cumsum(attention_mask, axis=1) - 1
                 position_ids[attention_mask == 0] = 1
-                if past_key_values:
-                    position_ids = position_ids[:, -input_ids.shape[1] :]
+            if past_len:
+                position_ids = position_ids[:, -inputs_embeds.shape[1] :]
 
             inputs["position_ids"] = position_ids
 
@@ -240,7 +251,7 @@ def __init__(
             self.lm_model,
             self.text_embdings_model,
             config=config,
-            deivce=device,
+            device=device,
             ov_config=ov_config,
             model_save_dir=model_save_dir,
             quantization_config=quantization_config,